Manuscript submitted to doi:10.3934/xx.xx.xx.xxAIMS’ JournalsVolume X, Number 0X, XX 200X pp. X–XX
DUURSMA’S REDUCED POLYNOMIAL
AZNIV KASPARIAN AND IVAN MARINOV
Azniv Kasparian, Ivan Marinov
Section of Algebra, Department of Mathematics and InformaticsKliment Ohridski University of Sofia
James Bouchier Blvd., Sofia 1164, Bulgaria
(Communicated by Marcus Greferath)
Abstract. The weight distribution {W(w)C }nw=0 of a linear code C ⊂ Fn
q isput in an explicit bijective correspondence with Duursma’s reduced polyno-
mial DC(t) ∈ Q[t] of C. We prove that the Riemann Hypothesis Analogue
for a linear code C requires the formal self-duality of C. Duursma’s reducedpolynomial DF (t) ∈ Z[t] of the function field F = Fq(X) of a curve X of genus
g over Fq is shown to provide a generating functionDF (t)
(1−t)(1−qt)=∞∑i=0Biti for
the numbers Bi of the effective divisors of degree i ≥ 0 of a virtual function
field of a curve of genus g − 1 over Fq .
Let Fq = ∪∞m=1Fqm be the algebraic closure of a finite field Fq andX/Fq ⊂ PN (Fq)be a smooth irreducible projective curve of genus g, defined over Fq. Denote byF = Fq(X) the function field of X over Fq and choose n different Fq-rational pointsP1, . . . , Pn ∈ X(Fq) := X ∩ PN (Fq). Suppose that G is an effective divisor of Fof degree 2g − 2 < degG = m < n, whose support is disjoint from the support ofD = P1 + . . . + Pn. The space L(G) := H0(X,OX(G)) of the global holomorphicsections of the line bundle, associated with G will be referred to as to the Riemann-Roch space of G. We put l(G) := dimFq L(G) and observe that the evaluationmap
ED : L(G) −→ Fnq ,
ED(f) = (f(P1), . . . , f(Pn)) for ∀f ∈ L(G)
is an Fq-linear embedding. Its image C := im(ED) = EDL(G) is known as analgebraic geometry code or Goppa code. The minimum distance of C is d(C) ≥n−m. The equality d(C) = n−m holds if and only if there exists a rational functionfo ∈ L(G), vanishing at exactly m of the points P1, . . . , Pn. For an arbitrary s ∈ N
2010 Mathematics Subject Classification. Primary: 94B27, 14G50; Secondary: 11 T71.Key words and phrases. Homogeneous weight enumerator of a linear code, Duursma’s zeta
polynomial and Duursma’s reduced polynomial of a linear code, Riemann Hypothesis Analogue
for linear codes, formally self-dual linear codes, Hasse-Weil polynomial and Duursma’s reduced
polynomial of a function field of one variable.Supported by Contract 015/9.04.2014 with the Scientific Foundation of the University of Sofia.
1
2 AZNIV KASPARIAN AND IVAN MARINOV
let Ns(F ) := |X(Fqs)| be the number of the Fqs-rational points of X. Then theformal power series
ZF (t) := exp
( ∞∑s=1
Ns(F )
sts
)is called the Hasse-Weil zeta function of F . It is well known (cf. Theorem 4.1.11from [6]) that
ZF (t) =LF (t)
(1− t)(1− qt)for a polynomial LF (t) ∈ Z[t] of degree 2g. We refer to LF (t) as to the Hasse-Weilpolynomial of F .
In [2], [3] Duursma introduces the genus of a linear code C ⊂ Fnq as the deviationg := n+ 1− k− d of its dimension k := dimFq C and minimum distance d from the
equality in Singleton bound. Let W(w)C be the number of the codewords c ∈ C of
weight d ≤ w ≤ n. Then
WC(x, y) := xn +
n∑w=d(C)
W(w)C xn−wyw
is called the homogeneous weight enumerator of C. Denote byMn,s(x, y) the MDS-weight enumerator of length n and minimum distance s. Put g⊥ for the genus ofthe dual code C⊥ of C and r := g + g⊥. In [2], [3] Duursma proves that thehomogeneous weight enumerator
WC(x, y) = a0Mn,d(x, y) + a1Mn,d+1(x, y) + . . .+ arMn,d+r(x, y). (1)
of an arbitrary linear code C ⊂ Fnq has uniquely determined coordinates a0, . . . , ar ∈Q with respect to the MDS-weight enumerators Mn,d+i(x, y), 0 ≤ i ≤ r. He refers
to PC(t) :=r∑i=0
aiti ∈ Q[t] as to the ζ-polynomial of C. The present note establishes
that the difference
WC(x, y)−Mn,n+1−k(x, y) = (q − 1)
r−2∑i=0
ci
(n
d+ i
)(x− y)n−d−iyd+i
of the homogeneous weight enumeratorWC(x, y) of C and the MDS-weight enumer-atorMn,n+1−k(x, y) of the same length n and dimension k as C has uniquely deter-mined coordinates c0, . . . , cr−2 ∈ Q with respect to (x− y)n−d−iyd+i, 0 ≤ i ≤ r− 2
(cf.Proposition 1). The polynomial DC(t) =r−2∑i=0
citi ∈ Q[t] is in a bijective corre-
spondence with PC(t) = (1 − t)(1 − qt)DC(t) + tg. Theorem 11.1 from Duursma’s[4] expresses the generating function ζC,j(t) = DC,j(t) + htg+j−1ZF (t) for the j-thsupport weights of C by a polynomial DC,j(t) and the Hasse-Weil ζ-function ZF (t)
of the function field F = Fq(Pj(Fq)) of the projective space Pj(Fq). In the case ofj = 1, Duursma’s DC,1(t) coincides with our DC(t) and that is why we call DC(t)Duursma’s reduced polynomial of C.
The classical Hasse-Weil Theorem establishes that all the roots of the Hasse-Weilpolynomial LF (t) ∈ Z[t] of the function field Fq(X) of a curve X of genus g over
Fq are on the circle S(
1√q
):{z ∈ C
∣∣∣ |z| = 1√q
}(cf. Theorem 4.2.3 form [6]).
Suppose that there is a complete set of representatives G1, . . . , Gh of the linearequivalence classes of the divisors of Fq(X) of degree 2g − 2 < degGi < n with
DUURSMA’S REDUCED POLYNOMIAL 3
Supp(Gi) ∩ Supp(D) = ∅ for ∀1 ≤ i ≤ n, D = P1 + . . .+ Pn. If Ci = EDL(Gi) arethe algebro-geometric Goppa codes, associated with these divisors, then accordingto Theorem 12.1 from Duursma’s [4], the ζ-polynomials of Ci are related by theequality ∑
tg−g(Ci)PCi(t) = LF (t).
to the Hasse-Weil polynomial LF (t) of F . Baring in mind this fact, Duursma saysthat a linear code C ⊂ Fnq satisfies the Riemann Hypothesis Analogue if all the
roots of its zeta polynomial PC(t) =r∑i=0
aiti ∈ Q[t] are on the circle S
(1√q
). Let
C be an Fq-linear code of dimension k and minimum distance d, which satisfiesthe Riemann Hypothesis Analogue. Proposition 2 shows that C is formally self-dual. Let us recall that C is formally self-dual if it has the same weight distribution
W(w)C = W(w)
C⊥ , ∀0 ≤ w ≤ n as its dual code C⊥ ⊂ Fnq . In the light of Duursma’sresults and our Proposition 1, the formal self-duality of C turns to be equivalent
to the functional equation PC(t) = PC
(1qt
)qgt2g for PC(t) and to the functional
equation DC(t) = DC
(1qt
)qg−1t2g−2 for DC(t). Proposition 3 from the present
note expresses explicitly the homogeneous weight enumeratorWC(x, y) of a formallyself-dual code C ⊂ Fnq by the lowest half of the coefficients of DC(t) or by the
numbers W(d)C , . . . ,W(k)
C of the codewords c ∈ C, whose weights are between theminimum distance d of C and the dimension k.
In [1] Dodunekov and Landgev introduce the near-MDS code C ⊂ Fnq as theones with quadratic zeta polynomial PC(t). Kim and Hyun’s article [5] provides anecessary and sufficient condition for a near-MDS code to satisfy the Riemann Hy-pothesis Analogue. By Theorem 3 from Duursma’s [3], the zeta polynomial PC(t) ofa formally self-dual code C ⊂ Fnq is of even degree. Our Proposition 4 is a necessaryand sufficient condition for a formally self-dual code C ⊂ Fnq with zeta polynomialPC(T ) of degPC(t) = 4 to be subject to the Riemann Hypothesis Analogue. Inanalogy with the classical Hasse-Weil Theorem, we intend to express the RiemannHypothesis Analogue for a linear code C ⊂ Fnq in terms of the coefficients of the
power series expansion of log[
PC(t)(1−t)(1−qt)
].
The last, third section is devoted to Duursma’s reduced polynomial DF (t) of thefunction field F = Fq(X) of a curve X/Fq ⊂ PN (Fq) of genus g over Fq. Corollary5.2 from Duursma’s [2] shows the existence ofDF (t). Explaining formula (10.1) from[4], he mentions that DF (t) accounts for the contribution of the special divisors ofF to the zeta function ZF (t). The present article establishes that DF (t) ∈ Z[t] isdetermined uniquely by its lowest g coefficients, which equal the numbers Ai of theeffective divisors of F of degree 0 ≤ i ≤ g − 1. Our Proposition 5 reveals that thezeta function
DF (t)
(1− t)(1− qt)=
∞∑i=0
Biti,
associated with DF (t) has the properties of a generating function for the numbersBi of the effective divisors of degree i ≥ 0 of a virtual function field of genus g − 1over Fq. There arises the following
Open Problem: To characterize the function fields F = Fq(X) of curves
X/Fq ⊂ PN (Fq) of genus g over Fq, for which there are curves Y/Fq ⊂ PM (Fq)
4 AZNIV KASPARIAN AND IVAN MARINOV
of genus g − 1, defined over Fq with Hasse-Weil zeta function
ZFq(Y )(t) =DF (t)
(1− t)(1− qt).
1. The homogeneous weight enumerator of an arbitrary code.
Proposition 1. Let C ⊂ Fnq be a linear code of dimension k = dimFq C, minimum
distance d and genus g = n + 1 − k − d ≥ 1, whose dual C⊥ ⊂ Fnq is of minimum
distance d⊥ and genus g⊥ = k + 1− d⊥ ≥ 1. If
DC(t) =
g+g⊥−2∑i=0
citi ∈ Q[t]
is Duursma’s reduced polynomial of C and Mn,n+1−k(x, y) is MDS-weight enumer-ator of length n, dimension k and minimum distance n+1−k, then the homogeneousweight enumerator of C is
WC(x, y) =Mn,n+1−k(x, y) + (q − 1)
g+g⊥−2∑i=0
ci
(n
d+ i
)(x− y)n−d−iyd+i. (2)
More precisely, Duursma’s reduced polynomial DC(t) =g+g⊥−2∑i=0
citi determines uni-
quely the weight distribution of C, according to
W(w)C = (q − 1)
(n
w
)w−d∑i=0
(−1)w−d−i(
w
d+ i
)ci for d ≤ w ≤ d+ g − 1, (3)
W(w)C =(q − 1)
(n
w
)min(w−d,n−d−d⊥)∑i=0
(−1)w−d−i(
w
d+ i
)ci
+
(n
w
)w−n−1+k∑j=0
(−1)j(w
j
)(qw−n+k−j − 1) for d+ g ≤ w ≤ n.
(4)
Conversely, for ∀0 ≤ i ≤ g + g⊥ − 2 the numbers W(d)C , . . . ,W(d+i)
C determine
uniquely the coefficient ci of Duursma’s reduced polynomial DC(t) =g+g⊥−2∑i=0
citi by
ci = (q − 1)−1(
n
d+ i
)−1 d+i∑w=d
(n− w
n− d− i
)W(w)C (5)
for 0 ≤ i ≤ g − 1,
ci = (q − 1)−1(
n
d+ i
)−1{d+g−1∑w=d
(n− w
n− d− i
)W(w)C
+
d+i∑w=d+g
(n− w
n− d− i
)W(w)C −
(n
w
)w−n−1+k∑j=0
(−1)j(w
j
)(qw−n+k−j − 1)
(6)
for g ≤ i ≤ g + g⊥ − 2.
DUURSMA’S REDUCED POLYNOMIAL 5
In particular,
(q − 1)
(n
d+ i
)ci ∈ Z
are integers for all 0 ≤ i ≤ g + g⊥ − 2.
The aforementioned formulae imply thatW(d)C , . . . ,W(d+g+g⊥−2)
C determine uniquelythe homogeneous weight enumerator WC(x, y) of C by the formula
WC(x, y) =
d+g+g⊥−2∑w=d
W(w)C λw(x, y) + Λ(x, y), (7)
with explicit polynomials
λw(x, y) :=
d+g+g⊥−2∑s=w
(n− wn− s
)(x− y)n−sys for d ≤ w ≤ d+ g + g⊥ − 2 (8)
and
Λ(x, y) :=Mn,n+1−k(x, y)−d+g+g⊥−2∑w=d+g
M(w)n,n+1−kλw(x, y). (9)
Proof. In the case of g = 0, note that C is an MDS-code andWC(x, y) =Mn,n+1−k(x, y).Form now on, we assume that g > 0 and put r := g + g⊥. According to Proposi-tion 9.2 from Duursma’s [2], the ζ-polynomials of C and C⊥ satisfy the functionalequation
PC⊥(t) = PC
(1
qt
)qgtg+g
⊥(10)
and PC(1) = PC⊥(1) = 1. Therefore PC
(1q
)= PC⊥(1)q−g =
(1q
)gand the
polynomial PC(t) − tg ∈ Q[t] vanishes at t = 1 and t = 1q . As a result, there is a
polynomial
Dc(t) :=PC(t)− tg
(1− t)(1− qt)=
r−2∑i=0
citi ∈ Q[t]. (11)
Making use of (1), let us express
WC(x, y) =Mn,d+g(x, y) +
r∑i=0
biMn,d+i(x, y)
by the coefficients of PC(t)− tg =r∑i=0
biti. The comparison of the coefficients of
PC(t)− tg = (1− t)(1− qt)DC(t). (12)
yields
bi = ci − (q + 1)ci−1 + qci−2 for ∀0 ≤ i ≤ rwith c−2 = c−1 = cr−1 = cr = 0. Therefore
WC(x, y) =Mn,d+g(x, y) +
r∑i=0
ciMn,d+i(x, y)
−(q + 1)
r∑i=0
ci−1Mn,d+i(x, y) + q
r∑i=0
ci−2Mn,d+i(x, y).
6 AZNIV KASPARIAN AND IVAN MARINOV
Setting j = i− 1, respectively, j = i− 2 in the last two sums, one obtains
WC(x, y) =Mn,d+g(x, y) +
r∑i=0
ciMn,d+i(x, y)
−(q + 1)
r−1∑j=−1
cjMn,d+j+1(x, y) + q
r−2∑j=−2
cjMn,d+j+2(x, y),
whereas
WC(x, y) =Mn,d+g(x, y)
+
r−2∑j=0
cj [Mn,d+j(x, y)− (q + 1)Mn,d+j+1(x, y) + qMn,d+j+2(x, y)].(13)
Let us put
Wn,d+j(x, y) :=Mn,d+j(x, y)− (q + 1)Mn,d+j+1(x, y) + qMn,d+j+2(x, y)
and recall that the MDS-weight enumerator of length n and minimum distance d+jequals
Mn,d+j(x, y) = xn +
n∑w=d+j
M(w)n,d+jx
n−wyw
with
M(w)n,d+j =
(n
w
)w−d−j∑i=0
(−1)i(w
i
)(qw+1−d−j−i − 1). (14)
Therefore
Wn,d+j(x, y) =M(d+j)n,d+jx
n−d−jyd+j + [M(d+j+1)n,d+j − (q + 1)M(d+j+1)
n,d+j+1]xn−d−j−1yd+j+1
+
n∑w=d+j+2
[M(w)n,d+j − (q + 1)M(w)
n,d+j+1 + qM(w)n,d+j+2]xn−wyw.
Making use of the MDS-weight distribution (14) and introducing
W(w)n,d+j :=M(w)
n,d+j − (q + 1)M(w)n,d+j+1 + qM(w)
n,d+j+2 for d+ j + 2 ≤ w ≤ n,
one expresses
Wn,d+j(x, y) =
(n
d+ j
)(q − 1)xn−d−jyd+j
−(
n
d+ j + 1
)(q − 1)(d+ j + 1)xn−d−j−1yd+j+1 +
n∑w=d+j+2
W(w)n,d+jx
n−wyw.
For any d+ j + 2 ≤ w ≤ n one has
W(w)n,d+j =
(n
w
)(w
d+ j
)(q − 1)(−1)w−d−j .
Baring in mind that (n
w
)(w
d+ j
)=
(n− d− jw − d− j
)(n
d+ j
),
DUURSMA’S REDUCED POLYNOMIAL 7
one obtains
Wn,d+j(x, y) =
(n
d+ j
)(q − 1)xn−d−jyd+j −
(n
d+ j + 1
)(q − 1)(d+ j + 1)xn−d−j−1yd+j+1+
+
n∑w=d+j+2
(n
d+ j
)(n− d− jw − d− j
)(q − 1)(−1)w−d−jxn−wyw.
Then by the means of
(d+ j + 1)
(n
d+ j + 1
)= (n− d− j)
(n
d+ j
),
one derives that
Wn,d+j(x, y) =
(n
d+ j
)(q − 1)
[xn−d−jyd+j − (n− d− j)xn−d−j−1yd+j+1+
+
n∑w=d+j+2
(−1)w−d−j(n− d− jw − d− j
)xn−wyw
.Introducing s := w − d− j, one expresses
n∑w=d+j+2
(−1)w−d−j(n− d− jw − d− j
)xn−wyw =
n−d−j∑s=2
(−1)s(n− d− j
s
)xn−d−j−syd+j+s
and concludes that
Wn,d+j(x, y) =
(n
d+ j
)(q − 1)(x− y)n−d−jyd+j . (15)
The equality Wn,n−k(x, y) =(nk
)(q − 1)(x − y)kyn−k is exactly the claim (c) of
Lemma 1 from Kim and Nyun’s work [5]. Plugging in (15) in (13) and bearing inmind that d+ g = n+ 1− k, one obtains (2).
In order to prove (3) and (4), let us put
VC(x, y) :=WC(x, y)−Mn,n+1−k(x, y)
and note that VC(x, y) =n∑
w=d
V(w)C xn−wyw with V(w)
C =W(w)C for d ≤ w ≤ n− k,
V(w)C =W(w)
C −M(w)n,n+1−k =W(w)
C −(n
w
)w−n−1+k∑i=0
(−1)i(w
i
)(qw−n+k−i − 1)
for d+ g = n+ 1− k ≤ w ≤ n. Making use of (2), one expresses
VC(x, y) = (q − 1)
g+g⊥−2∑i=0
ci
(n
d+ i
) n−d−i∑s=0
(n− d− i
s
)(−1)n−d−i−sxsyn−s
= (q − 1)
n−d∑s=0
min(n−d−s,g+g⊥−2)∑i=0
ci
(n
d+ i
)(n− d− i
s
)(−1)n−d−i−s
xsyn−s,after changing the summation order. Setting w := n− s, one obtains
VC(x, y) = (q−1)
n∑w=d
min(w−d,n−d−d⊥)∑i=0
ci
(n
d+ i
)(n− d− in− w
)(−1)w−d−i
xn−wyw.
8 AZNIV KASPARIAN AND IVAN MARINOV
Then (n
d+ i
)(n− d− in− w
)=
(n
w
)(w
d+ i
),
allows to concludes that
V(w)C = (q − 1)
(n
w
)min(w−d,n−d−d⊥)∑i=0
ci
(w
d+ i
)(−1)w−d−i for ∀d ≤ w ≤ n,
which proves (3), (4).Towards (5), (6), let us introduce z := x− y and express (2) in the form
VC(y + z, y) = (q − 1)
g+g⊥−2∑i=0
ci
(n
d+ i
)zn−d−iyd+i. (16)
On the other hand,
VC(y + z, y) =
n∑w=d
V(w)C (y + z)n−wyw
=
n∑w=d
n−w∑s=0
(n− ws
)V(w)C yn−szs =
n−d∑s=0
[n−s∑w=d
(n− ws
)V(w)C
]yn−szs,
after changing the summation order. Comparing the coefficients of yd+izn−d−i inthe left and right hand side of (16), one obtains
d+i∑w=d
(n− w
n− d− i
)V(w)C = (q − 1)ci
(n
d+ i
),
whereas
ci = (q − 1)−1(
n
d+ i
)−1 d+i∑w=d
(n− w
n− d− i
)V(w)C .
Combining with (14), one justifies (5) and (6). These formulae imply also that(q − 1)
(nd+i
)ci ∈ Z are integers for all 0 ≤ i ≤ g + g⊥ − 2.
The substitution by (5), (6), (14) in (2) yields
WC(x, y) =Mn,n+1−k(x, y) +
g+g⊥−2∑i=0
d+i∑w=d
(n− w
n− d− i
)W(w)C (x− y)n−d−iyd+i
−g+g⊥−2∑i=g
d+i∑w=d+g
(n− w
n− d− i
)M(w)
n,n+1−k(x− y)n−d−iyd+i.
One exchanges the summation order in the double sums towards
WC(x, y) =Mn,n+1−k(x, y) +
d+g+g⊥−2∑w=d
W(w)C
g+g⊥−2∑i=w−d
(n− w
n− d− i
)(x− y)n−d−iyd+i
−d+g+g⊥−2∑w=d+g
M(w)n,n+1−k
g+g⊥−2∑i=w−d
(n− w
n− d− i
)(x− y)n−d−iyd+i.
Introducing s := d+ i, one obtains (7) with (8) and (9).
DUURSMA’S REDUCED POLYNOMIAL 9
Comparing the coefficients of xn−dyd in the left and right hand sides of (2), one
obtains W(d)C = (q − 1)
(nd
)c0 for a linear code C of genus g ≥ 1. We claim that
c0 < 1. To this end, note that for any d-tuple {i1, . . . , id} ⊂ {1, . . . , n}, supportinga word c ∈ C of weight d there are exactly q − 1 words c′ ∈ C with Supp(c′) =Supp(c) = {i1, . . . , id}. That is due to the fact that the columns Hi1 , . . . ,Hid of anarbitrary parity check matrix H of C are of rank d − 1 and there are no words ofweight ≤ d−1 in the right null space of the matrix (Hi1 . . . Hid). If ν is the number
of the supports of the words of C of weight d then ν(q − 1) =W(d)C , whereas
c0 =W(d)C
(q − 1)(nd
) =ν(nd
) ≤ 1.
If we assume that c0 = 1 then any d-tuple of columns of H is linearly dependent.Bearing in mind that rkH = n− k, one concludes that d > n− k. Combining withSingleton Bound d ≤ n − k + 1, one obtains d = n − k + 1. That contradicts theassumption that C is not an MDS-code and proves that c0 < 1 for any Fq-linearcode C ⊂ Fnq of genus g ≥ 1. Note that c0 can be interpreted as the probability fora d-tuple to support a word of weight d from C.
2. The Riemann Hypothesis Analogue and the formal self-duality of alinear code. Recall that a linear code C ⊂ Fnq with dual C⊥ ⊂ Fnq is formally
self-dual if C and C⊥ have one and a same number W(w)C = W(w)
C⊥ of codewordsof weight 0 ≤ w ≤ n. Let us mention some trivial consequences of the formalself-duality of C. First of all, C and C⊥ have one and a same minimum distanced = d(C) = d(C⊥) = d⊥. Further, C and C⊥ have one and a same cardinality
qdimC =
n∑w=0
W(w)C =
n∑w=0
W(w)C = qdimC⊥
,
so that k = dimC = dimC⊥ = k⊥ and the length n = k + k⊥ = 2k is an even
integer. The genera g = k + 1 − d = g⊥ also coincide. Let PC(t) =2g∑i=0
aiti and
PC⊥ =2g∑i=0
a⊥i ti be the zeta polynomials of C, respectively, of C⊥. The consecutive
comparison of the coefficients of xn−dyd, xn−d−1yd+1, . . . , xn−d−2gyd+2g from thehomogeneous polynomial
a0M2k,d(x, y) + a1M2k,d+1(x, y) + . . .+ a2gM2k,d+2g(x, y) =WC(x, y)
=WC⊥(x, y) = a⊥0M2k,d(x, y) + a⊥1M2k,d+1(x, y) + . . .+ a⊥2gM2k,d+2g(x, y)
in x, y yields ai = a⊥i for ∀0 ≤ i ≤ 2g. It is clear that ai = a⊥i for ∀0 ≤ i ≤ 2g sufficesfor WC(x, y) = WC⊥(x, y), so that the formal self-duality of C is tantamount tothe coincidence PC(t) = PC⊥(t) of the zeta polynomials of C and C⊥. Duursma
has shown in Proposition 9.2 from [2] that Mac Williams identities for W(w)C and
W(w)
C⊥ are equivalent to the functional equation (10) for the zeta polynomials PC(t),
PC⊥(t) of C,C⊥ ⊂ Fnq with genera g, g⊥. Thus, an Fq-linear code C ⊂ Fnq is formallyself-dual if and only if its zeta polynomial PC(t) satisfies the functional equation
PC(t) = PC
(1
qt
)qgt2g (17)
of the Hasse-Weil polynomial of the function field of a curve of genus g over Fq.
10 AZNIV KASPARIAN AND IVAN MARINOV
Proposition 2. If a linear code C ⊂ Fnq satisfies the Riemann Hypothesis Analoguethen C is formally self-dual, i.e., the zeta polynomial PC(t) of C is subject to thefunctional equation (17) of the Hasse-Weil polynomial of the function field of acurve of genus g over Fq.
Proof. Let us assume that PC(t) of degree r := g + g⊥ satisfies the Riemann Hy-pothesis Analogue, i.e.,
PC(t) = ar
r∏j=1
(t− αj) ∈ Q[t]
for some αj ∈ C with |αj | = 1√q for all 1 ≤ j ≤ r. If αj is a real root of PC(t) then
αj = ε√q with ε = ±1. We claim that in the case of an even degree r = 2m, the
zeta polynomial PC(t) is of the form
PC(t) = a2m
m∏i=1
(t− αi)(t− αi) (18)
or of the form
PC(t) = a2m
(t2 − 1
q
)m−1∏i=1
(t− αi)(t− αi), (19)
while for an odd degree r = 2m+ 1 one has
PC(t) = a2m+1
(t− ε√q
) m∏i=1
(t− αi)(t− αi) (20)
for some ε ∈ {±1}. Indeed, if αi ∈ C\R is a complex, non-real root of PC(t) ∈ Q[t] ⊂R[t] then αi 6= αi is also a root of PC(t) and PC(t) is divisible by (t−αi)(t−αi). If
PC(t) = 0 has three real roots α1, α2, α3 ∈{
1√q ,−
1√q
}, then at least two of them
coincide. For α1 = α2 = ε√q one has (t − α1)(t − α2) = (t − α1)(t − α1). Thus,
PC(t) has at most two real roots, which are not complex conjugate (or, equivalently,equal) to each other and PC(t) is of the form (18), (19) or (20).
If PC(t) is of the form (18), then PC(t) = a2mm∏i=1
(t2 − 2Re(αi) + 1
q
)and (10)
reads as
PC⊥(t) = a2m
[m∏i=1
(1
q− 2Re(αi)t+ t2
)]qg−m = PC(t)qg−m, (21)
after multiplying each of the factors 1q2t2 −
2Re(αi)qt + 1
q by qt2. If DC(t) is Duursma’s
reduced polynomial of C and DC⊥(t) is Duursma’s reduced polynomial of C⊥, then
(1−t)(1−qt)DC⊥(t)+tg⊥
= PC⊥(t) = PC(t)qg−m = (1−t)(1−qt)qg−mDC(t)+qg−mtg
implies that
(1− t)(1− qt)[DC⊥(t)− qg−mDC(t)] = qg−mtg − tg⊥.
Plugging in t = 1, one concludes that qg−m = 1, whereas g = m. As a result,g + g⊥ = 2m = 2g specifies that g = g⊥ and (21) yields PC(t) = PC⊥(t), which isequivalent to the formal self-duality of C.
DUURSMA’S REDUCED POLYNOMIAL 11
If PC(t) is of the form (19) then (10) provides
PC⊥(t) = a2m
(1
q− t2
)[m−1∏i=1
(1
q− 2Re(αi)t+ t2
)]qg−m = −PC(t)qg−m. (22)
Expressing by Duursma’s reduced polynomials DC(t), DC⊥(t), one obtains
(1− t)(1− qt)DC⊥(t) + tg⊥
= PC⊥(t) =
−PC(t)qg−m = −(1− t)(1− qt)qg−mDC(t)− qg−mtg,
whereas
(1− t)(1− qt)[DC⊥(t) + qg−mDC(t)] = −tg⊥− qg−mtg.
The substitution t = 1 in the last equality of polynomials yields −1 − qg−m = 0,which is an absurd, justifying that a zeta polynomial PC(t), subject to the RiemannHypothesis Analogue cannot be of the form (19).
If PC(t) is of odd degree 2m+ 1, then (20) and (10) yield
PC⊥(t) = −ε√qa2m+1
(t− ε√q
)[ m∏i=1
(1
q− 2Re(αi)t+ t2
)]qg−m−1
= −ε√qPC(t)qg−m−1
after multiplying 1qt −
ε√q by − ε√
q qt and each 1q2t2 −
2Re(αi)qt + 1
q by qt2. Expressing
by Duursma’s reduced polynomials
(1− t)(1− qt)DC⊥(t) + tg⊥
= PC⊥(t) = −εqg−m− 12PC(t)
= −εqg−m− 12 (1− t)(1− qt)DC(t)− εqg−m− 1
2 tg,
one obtains
(1− t)(1− qt)[DC⊥(t) + εqg−m−
12DC(t)
]= −tg
⊥− εqg−m− 1
2 tg.
The substitution t = 1 implies −1 − εqg−m− 12 = 0, which is an absurd, as far as
qx = 1 if and only if x = 0, while g −m− 12 cannot vanish for integers g,m. Thus,
none zeta polynomial of odd degree satisfies the Riemann Hypothesis Analogue.
Proposition 3. The following conditions are equivalent for a linear code C ⊂ Fnq :(i) C is formally self-dual, i.e., the zeta polynomial PC(t) of C satisfies the
functional equation
PC(t) = PC
(1
qt
)qgt2g
of the Hasse-Weil polynomial of the function field of a curve of genus g over Fq;
(ii) Duursma’s reduced polynomial DC(t) =g+g⊥−2∑i=0
citi satisfies the functional
equation
DC(t) = DC
(1
qt
)qg−1t2g−2 (23)
of the Hasse-Weil polynomial of the function field of a curve of genus g−1 over Fq;
12 AZNIV KASPARIAN AND IVAN MARINOV
(iii) the coefficients of Duursma’s reduced polynomial DC(t) =g+g⊥−2∑i=0
citi of C
satisfy the equalities
cg−1+i = qicg−1−i for ∀1 ≤ i ≤ g − 1; (24)
(iv) the dual code C⊥ ⊂ Fnq of C has dimension dimFq C⊥ = dimFq C = k, genus
g(C⊥) = g(C) = g and the homogeneous weight enumerator of C is
WC(x, y) =M2k,k+1(x, y) +
g−1∑j=0
cg−1−jwj(x, y), (25)
where
wj(x, y) := (q − 1)
(2k
k + j
)[(x− y)k+jyk−j + qj(x− y)k−jyk+j
](26)
for 1 ≤ j ≤ g − 1.
w0(x, y) := (q − 1)
(2k
k
)(x− y)kyk. (27)
(v) the dual code C⊥ ⊂ Fnq of C has dimension dimFq C⊥ = dimFq C = k, genus
g(C⊥) = g(C) = g and the homogeneous weight enumerator
WC(x, y) =M2k,k+1(x, y) +
k−1∑w=d
W(w)C ϕw(x, y) +W(k)
C (x− y)kyk (28)
with
ϕw(x, y) :=
k−1∑s=w
(2k − ws− w
)[(x− y)2k−sys + qk−s(x− y)sy2k−s
]+
(2k − wk
)(x−y)kyk
(29)for d ≤ w ≤ k − 1, so that C can be obtained from an MDS-code of the samelength 2k and dimension k by removing and adjoining appropriate words, depending
explicitly on the numbers W(d)C ,W(d+1)
C , . . . ,W(k)C of the codeword of C of weight
≤ k = dimFq C.
Proof. Towards (i)⇒ (ii), one substitutes by PC(t) = (1− t)(1− qt)DC(t) + tg in(17), in order to obtain
(1− t)(1− qt)DC(t) + tg = (qt− 1)(t− 1)
[DC
(1
qt
)qg−1t2g−2
]+ tg,
whereas (23).Conversely, (ii)⇒ (i) is justified by
PC(t) = (1− t)(1− qt)DC(t) + tg =
= (t− 1)(qt− 1)
[DC
(1
qt
)qg−1t2g−2
]+ tg
=
[(1− 1
t
)t
] [(1− 1
qt
)qt
] [DC
(1
qt
)qg−1t2g−2
]+qgt2g
qgtg
=
[(1− q
qt
)(1− 1
qt
)DC
(1
qt
)+
1
(qt)g
]qgt2g = PC
(1
qt
)qgt2g.
That proves the equivalence (i)⇔ (ii).
DUURSMA’S REDUCED POLYNOMIAL 13
Towards (ii)⇔ (iii), note that the functional equation of DC(t) reads as
2g−2∑i=0
citi = DC(t) = DC
(1
qt
)qg−1t2g−2 =
(2g−2∑i=0
ciqiti
)qg−1t2g−2
=
2g−2∑i=0
ciqg−1−it2g−2−i =
2g−2∑j=0
c2g−2−jq−g+1+jtj .
Comparing the coefficients of the left-most and the right-most side, one expressesthe formal self-duality of C by the relations
cj = q−g+1+jc2g−2−j for ∀0 ≤ j ≤ 2g − 2.
Let i := g − 1− j, in order to transform the above conditions to
cg−1+i = qicg−1−i for ∀ − g + 1 ≤ i ≤ g − 1. (30)
For any −g + 1 ≤ i ≤ −1 note that cg−1+i = qicg−1−i is equivalent to cg−1−i =q−icg−1+i and follows from (30) with 1 ≤ −i ≤ g − 1. In the case of i = 0, (30)holds trivially and (30) amounts to (24). That proves the equivalence of (ii) with(iii).
Towards (iii)⇒ (iv), one introduces a new variable z := x− y and expresses (2)in the form
VC(y + z, y) :=WC(y + z, y)−M2k,k+1(y + z, y) = (q − 1)
2g−2∑i=0
ci
(2k
d+ i
)yd+iz2k−d−i
= (q − 1)
g−1∑i=0
ci
(2k
d+ i
)yd+iz2k−d−i + (q − 1)
2g−2∑i=g
ci
(2k
d+ i
)yd+iz2k−d−i.
Let us change the summation index of the first sum to 0 ≤ j := g − 1− i ≤ g − 1,put 1 ≤ j := i− g + 1 ≤ g − 1 in the second sum and make use of d+ g = k + 1, inorder to obtain
VC(y + z, y)
= (q − 1)
g−1∑j=0
cg−1−j
(2k
k − j
)yk−jzk+j + (q − 1)
g−1∑j=1
cj+g−1
(2k
k + j
)yk+jzk−j .
(31)
Extracting the term with j = 0 from the first sum, one expresses
VC(y + z, y) = (q − 1)cg−1
(2k
k
)ykzk
+
g−1∑j=1
(q − 1)
(2k
k + j
)[cg−1−jy
k−jzk+j + cg−1+jyk+jzk−j
] (32)
for an arbitrary Fq-linear code C ⊂ Fnq . If C is formally self-dual, then plugging inby (24) in (32) and making use of (26), (27), one gets
VC(y + z, y) =
g−1∑j=0
cg−1−jwj(y + z, y).
Substituting z := x− y and VC(x, y) :=WC(x, y)−M2k,k+1(x, y), one derives theequality (25) for the homogeneous weight enumerator of a formally self-dual linearcode C ⊂ F2k
q .
14 AZNIV KASPARIAN AND IVAN MARINOV
In order to justify that (iv) suffices for the formal self-duality of C, we use that(25) with (26) and (27) is equivalent to
VC(y + z, y) =
g−1∑j=1
cg−1−j(q − 1)
(2k
k + j
)yk−jzk+j
+cg−1(q − 1)
(2k
k
)ykzk +
g−1∑j=1
cg−1−j(q − 1)
(2k
k + j
)yk+jzk−j
(33)
Comparing the coefficients of yk+jzk−j with 1 ≤ j ≤ g − 1 from (32) and (33), oneconcludes that
cg−1+j = cg−1−jqj for ∀1 ≤ j ≤ g − 1.
These are exactly the relations (24) and imply the formal self-duality of C.
Towards (iv)⇔ (v), it suffices to put E(x, y) :=g−1∑j=0
cg−1−jwj(x, y) and to derive
that E(x, y) =k−1∑w=d
W(w)C ϕw(x, y) +W(k)
C (x − y)kyk. More precisely, introducing
i := g − 1− j, one expresses
E(x, y) =
g−2∑i=0
ci(q − 1)
(2k
d+ i
)[(x− y)2k−d−iyd+i + qg−1−i(x− y)d+iy2k−d−i
]+cg−1(q − 1)
(2k
k
)(x− y)kyk.
Plugging in by (5) and exchanging the summation order, one gets
E(x, y) =
k−1∑w=d
g−2∑i=w−d
(2k − wd+ i− w
)W(w)C [(x− y)2k−d−iyd+i + qg−1−i(x− y)d+iy2k−d−i]
+
k∑w=d
(2k − wk
)W(w)C (x− y)kyk.
Introducing s := d+ i and extracting W(w)C as coefficients, one obtains
E(x, y) =
k−1∑w=d
W(w)C ϕw(x, y) +W(k)
C (x− y)kyk.
Let C ⊂ Fnq be an Fq-linear code of genus g, whose dual C⊥ ⊂ Fnq is of genus
g⊥. In [1], Dodunekov and Landgev introduce the near-MDS linear codes C as theones with zeta polynomial PC(t) ∈ Q[t] of degree degPC(t) := g + g⊥ = 2. Thus,C is a near-MDS code if and only if it has constant Duursma’s reduced polynomialDC(t) = c0 ∈ Q. Kim an Hyun prove in [5]) that a near-MDS code C satisfies theRiemann Hypothesis Analogue exactly when
1
(√q + 1)2
≤ c0 ≤1
(√q − 1)2
.
The next proposition characterizes the formally-self-dual codes C ⊂ Fnq of genus2, which satisfy the Riemann Hypothesis Analogue. By Proposition 3 (iii), C is
DUURSMA’S REDUCED POLYNOMIAL 15
a formally self-dual linear code of genus 2 exactly when its Duursma’s reducedpolynomial is
DC(t) = c0 + c1t+ qc0t2
for some c0, c1 ∈ Q, 0 < c0 < 1.
Proposition 4. A formally self-dual linear code C ⊂ F2kq with a quadratic Du-
ursma’s reduced polynomial DC(t) = c0 + c1t+ qc0t2 ∈ Q[t], 0 < c0 < 1 satisfies the
Riemann Hypothesis Analogue if and only if
[(q + 1)c0 + c1]2 ≥ 4c0, (34)
q − 4√q + 1 ≤ c1
c0≤ q + 4
√q + 1, (35)
c1 ≤ min
(1
(√q − 1)2
− 2√qc0,
1
(√q + 1)2
+ 2√qc0
). (36)
Proof. According to (18) from the proof of Proposition 2, the zeta polynomial
PC(t) = (1− t)(1− qt)(qc0t2 + c1t+ c0) + t2
satisfies the Riemann Hypothesis Analogue if and only if there exist ϕ,ψ ∈ [0, 2π)with
PC(t) = q2c0
(t− eiϕ√q
)(t− e−iϕ
√q
)(t− eiψ√q
)(t− e−iψ
√q
).
Comparing the coefficients of t and t2 from PC(t), one expresses this condition bythe equalities
c1 − (q + 1)c0 = −2√qc0[cos(ϕ) + cos(ψ)],
1 + 2qc0 − (q + 1)c1 = 2qc0[1 + 2 cos(ϕ) cos(ψ)].
These are equivalent to
cos(ϕ) + cos(ψ) =(q + 1)c0 − c1
2√qc0
and
cos(ϕ) cos(ψ) =1− (q + 1)c1
4qc0.
In other words, the quadratic equation
f(t) := t2 +c1 − (q + 1)c0
2√qc0
t+1− (q + 1)c1
4qc0∈ Q[t]
has roots −1 ≤ t1 = cos(ϕ) ≤ t2 = cos(ψ) ≤ 1. This, in turn, holds exactly whenthe discriminant
D(f) =
[c1 − (q + 1)c0
2√qc0
]2− 4[1− (q + 1)c1]
4qc0≥ 0 (37)
is non-negative, the vertex
− 1 ≤ (q + 1)c0 − c14√qc0
≤ 1 (38)
belongs to the segment [−1, 1] and the values of f(t) at the ends of this segmentare non-negative,
f(1) ≥ 0, f(−1) ≥ 0. (39)
16 AZNIV KASPARIAN AND IVAN MARINOV
The equivalence of (37) to (34) is straightforward. Since C is of minimum distance
d = k− 1 and W(k−1)C = (q− 1)
(2kk−1)c0 ∈ N, the constant term c0 > 0 of DC(t) is a
positive rational number and one can multiply (38) by −4√qc0 < 0, add (q + 1)c0
to all the terms and rewrite it in the form
(q − 4√q + 1)c0 ≤ c1 ≤ (q + 4
√q + 1)c0.
Making use of c0 > 0, one observes that the above inequalities are tantamount to(35). Finally,
4qc0f(1) = 4qc0+2√q[c1−(q+1)c0]+1−(q+1)c1 = (−c1−2
√qc0)(
√q−1)2+1 ≥ 0
and
4qc0f(−1) = 4qc0−2√q[c1−(q+1)c0]+1−(q+1)c1 = (2
√qc0−c1)(
√q+1)2+1 ≥ 0
can be expressed as (36).
3. Duursma’s reduced polynomial of a function field. Let F = Fq(X) bethe function field of a curve X of genus g over Fq and hg := h(F ) be the classnumber of F , i.e., the number of the linear equivalence classes of the divisors ofF of degree 0. The present section introduces an additive decomposition of theHasse-Weil polynomial LF (t) ∈ Z[t] of F , which associates to F a sequence {hi}g−1i=1
of virtual class numbers hi of function fields of curves of genus i over Fq.
Lemma 3.1. The following conditions are equivalent for a polynomial Lg(t) ∈ Q[t]of degree degLg(t) = 2g:
(i) Lg(t) satisfies the functional equation
Lg(t) = Lg
(1
qt
)qgt2g
of the Hasse-Weil polynomial of the function field of a curve of genus g over Fq;
(ii) Lg−1(t) :=Lg(t)− Lg(1)tg
(1− t)(1− qt)is a polynomial with rational coefficients of degree 2g − 2, satisfying the functionalequation
Lg−1(t) = Lg−1
(1
qt
)qg−1t2g−2
of the Hasse-Weil polynomial of the function field of a curve of genus g−1 over Fq;
(iii) Lg(t) =
g∑i=0
hiti(1−t)g−i(1−qt)g−i
for some rational numbers hi ∈ Q.
Proof. Towards (i)⇒ (ii), let us note that the polynomial Mg(t) := Lg(t)−Lg(1)tg
vanishes at t = 1, so that it is divisible by 1− t. Further,
Mg(t) = Lg(t)− Lg(1)tg =
[Lg
(1
qt
)− Lg(1)
qgtg
]qgt2g = Mg
(1
qt
)qgt2g
satisfies the functional equation of the Hasse-Weil polynomial of the function field
of a curve of genus g over Fq. In particular, Mg
(1q
)= Mg(1) q
g
q2g = 0 and Mg(t) is
DUURSMA’S REDUCED POLYNOMIAL 17
divisible by the linear polynomial q(
1q − t
)= 1 − qt, which is relatively prime to
1− t in Q[t]. As a result,
Lg−1(t) :=Mg(t)
(1− t)(1− qt)∈ Q[t]
is a polynomial of degree degLg−1(t) = 2g − 2. Straightforwardly,
Lg−1
(1
qt
)qg−1t2g−2 =
[Mg
(1
qt
):
(1− 1
qt
)(1− 1
t
)]qg−1t2g−2
=Mg(t)
qt2:
(qt− 1)(t− 1)
qt2=
Mg(t)
(1− t)(1− qt)= Lg−1(t)
satisfies the functional equation of the Hasse-Weil polynomial of the function fieldof a curve of genus g − 1 over Fq.
The implication (ii) ⇒ (i) follows from the functional equation of Lg−1(t), ap-plied to Lg(t) = (1− t)(1− qt)Lg−1(t) + Lg(1)tg. Namely,
Lg
(1
qt
)qgt2g
=
[(1− 1
qt
)qt
] [(1− 1
t
)t
] [Lg−1
(1
qt
)qg−1t2g−2
]+Lg(1)
qgtgqgt2g
= (qt− 1)(t− 1)Lg−1(t) + Lg(1)tg
= (1− t)(1− qt)Lg−1(t) + Lg(1)tg = Lg(t).
We derive (i) ⇒ (iii) by an induction on g, making use of (ii). More precisely,
for g = 1 one has L0(t) := L1(t)−L1(1)t(1−t)(1−qt) ∈ Q[t] of degree degL0(t) = 0 or L0 ∈ Q.
Then
L1(t) = (1− t)(1− qt)L0 + L1(1)t =
1∑i=0
hiti(1− t)1−i(1− qt)1−i
with h0 := L0 ∈ Q and h1 := L1(1) ∈ Q. In the general case, (ii) provides apolynomial
Lg−1(t) :=Lg(t)− Lg(1)tg
(1− t)(1− qt),
subject to the functional equation
Lg−1(t) = Lg−1
(1
qt
)qg−1t2g−2
of the Hasse-Weil polynomial of the function field of a curve of genus g− 1 over Fq.By the inductional hypothesis, there exist h′i ∈ Q, 0 ≤ i ≤ g − 1 with
Lg−1(t) =
g−1∑i=0
h′iti(1− t)g−1−i(1− qt)g−1−i.
Then
Lg(t) = (1− t)(1− qt)Lg−1(t) + Lg(1)tg =
g∑i=0
hiti(1− t)g−i(1− qt)g−i
with hi := h′i ∈ Q for 0 ≤ i ≤ g − 1 and hg := Lg(1) ∈ Q justifies (i)⇒ (iii).
18 AZNIV KASPARIAN AND IVAN MARINOV
Towards (iii) ⇒ (i), let us assume that Lg(t) =g∑i=0
hiti(1 − t)g−i(1 − qt)g−i.
Then
L
(1
qt
)qgt2g =
[g∑i=0
hiqiti
(1− 1
qt
)g−i(1− 1
t
)g−i]qgt2g
=
g∑i=0
[hiqiti
qit2i] [(
1− 1
qt
)qt
]g−i [(1− 1
t
)t
]g−i=
g∑i=0
hiti(qt− 1)g−i(t− 1)g−i = Lg(t)
satisfies the functional equation of the Hasse-Weil polynomial of the function fieldof a curve of genus g over Fq.
Proposition 5. Let F = Fq(X) be the function field of a smooth irreducible curve
X/Fq ⊂ PN (Fq) of genus g, defined over Fq, with h(F ) linear equivalence classesof divisors of degree 0, Ai effective divisors of degree i ≥ 0, Hasse-Weil polyno-mial LF (t) ∈ Q[t] and Duursma’s reduced polynomial DF (t) ∈ Q[t], defined by theequality
LF (t) = (1− t)(1− qt)DF (t) + h(F )tg.
Then:
(i) DF (t) =g−2∑i=0
Ai(ti + qg−1−it2g−2−i) + Ag−1tg−1 ∈ Z[t] is a polynomial with
integral coefficients, which is uniquely determined by A0 = 1,A1, . . . ,Ag−1;(ii) the equality
DF (t)
(1− t)(1− qt)=
∞∑i=0
Biti (40)
of formal power series of t holds for
Bi =
i∑j=0
Aj(qi−j+1 − 1
q − 1
)(41)
for 0 ≤ i ≤ g − 1,
Bi =
g−1∑j=0
Aj(qi−j+1 − 1
q − 1
)+
i∑j=g
A2g−2−j
(qi−g+2 − qj−g+1
q − 1
)(42)
for g ≤ i ≤ 2g − 3,
Bi = DF (1)
(qi−g+2 − 1
q − 1
)(43)
for i ≥ 2g − 2;(iii) the natural numbers Bi, i ≥ 0 from (ii) satisfy the relations
Bi = qi−g+2B2g−4−i +DF (1)
(qi−g+2 − 1
q − 1
)for ∀g − 1 ≤ i ≤ 2g − 4; (44)
Bi = DF (1)
(qi−g+2 − 1
q − 1
)for ∀i ≥ 2g − 3. (45)
DUURSMA’S REDUCED POLYNOMIAL 19
(iv) the number h(F ) of the linear equivalence classes of the divisors of F ofdegree 0 satisfies the inequilities
(√q − 1)2g ≤ h(F ) ≤ (
√q + 1)2g
Proof. (i) By Theorem 4.1.6. (ii) and Theorem 4.1.11 from [6], the Hasse-Weil zetafunction of F is the generating function
ZF (t) =LF (t)
(1− t)(1− qt)=
∞∑j=0
Ajtj
of the sequence {Ai}∞i=0. According to Lemma 3.1 and LF (1) = h(F ),
DF (t) :=LF (t)− h(F )tg
(1− t)(1− qt)
is a polynomial of degDF (t) = 2g − 2, subject to the functional equation of theHasse-Weil polynomial of the function field of a curve of genus g− 1 over Fq. Thus,
ZF (t) = DF (t) +h(F )tg
(1− t)(1− qt)=
∞∑j=0
Ajtj . (46)
Let l(G) is the dimension of the space H0(X,OX(G)) of the global holomorphicsections of the line bundle OX(G) → X, associated with a divisor G ∈ Div(F ).Riemann-Roch Theorem asserts that
l(G) = l(KX −G) + deg(G)− g + 1
for a canonical divisor KX of X. For any j ≥ g − 1, suppose that G1, . . . , Gh(F ) ∈Div(F ) is a complete set of representatives of the linear equivalence classes of thedivisors of F of degree j. Then
Aj =
h(F )∑ν=1
ql(Gν) − 1
q − 1= qj−g+1
h(F )∑ν=1
(ql(KY −Gν) − 1
q − 1
)+ h(F )
(qj−g+1 − 1
q − 1
)(47)
for g ≤ j ≤ 2g − 2 and
Aj = h(F )
(qj−g+1 − 1
q − 1
)for ∀j ≥ 2g − 1. (48)
Note that KY −G1, . . . ,KY −Gh(F ) is a complete set of representatives of the linearequivalence classes of the divisors of F of degree 2g − 2− j, so that
A2g−2−j =
h(F )∑ν=1
ql(KY −Gν) − 1
q − 1. (49)
Plugging in by (49) in (47), one obtains
Aj = qj−g+1A2g−2−j + h(F )
(qj−g+1 − 1
q − 1
)for g ≤ j ≤ 2g − 2, (50)
whereas
ZF (t) =
g−1∑j=0
Ajtj +
2g−2∑j=g
qj−g+1A2g−2−jtj + h(F )
∞∑j=g
(qj−g+1 − 1
q − 1
)tj ,
20 AZNIV KASPARIAN AND IVAN MARINOV
Putting i := 2g − 2 − j in the second sum and i := j − g in the third sum, oneexpresses
ZF (t) =
g−2∑i=0
Ai(ti + qg−1−it2g−2−i) +Ag−1tg−1
+h(F )
[qtg
q − 1
( ∞∑i=0
qiti
)− tg
q − 1
( ∞∑i=0
ti
)],
Summing up the geometric progressions
∞∑i=0
qiti =1
1− qt,
∞∑i=0
ti =1
1− t,
one derives
ZF (t) =
g−2∑i=0
Ai(ti + qg−1−it2g−2−i) +Ag−1tg−1 + h(F )tg
(1− t)(1− qt),
whereas
DF (t) =
g−2∑i=0
Ai(ti + qg−1−it2g−2−i) +Ag−1tg−1.
In particular, DF (t) ∈ Z[t] has integral coefficients.(ii) Let us expand
1
1− t=
∞∑i=0
ti,1
1− qt=
∞∑i=0
qiti
as sums of geometric progressions and note that
1
(1− t)(1− qt)=
∞∑i=0
(1 + q + . . .+ qi)ti =
∞∑i=0
(qi+1 − 1
q − 1
)ti.
Then represent Duursma’s reduced polynomial in the form
DF (t) =
g−1∑j=0
Ajtj +
2g−2∑j=g
A2g−2−jqj−g+1tj . (51)
Now, the comparison of the coefficients of ti, i ≥ 0 from the left hand side and theright hand side of (40) provides (41), (42) and
Bi =
g−1∑j=0
Aj(qi−j+1 − 1
q − 1
)+
2g−2∑j=g
A2g−2−jqj−g+1
(qi−j+1 − 1
q − 1
)for i ≥ 2g − 2.
The last formula can be expressed in the form
Bi =qi+1
q − 1
q−1∑j=0
Ajq−j +
2g−2∑j=g
A2g−2−jqj−g+1q−j
− 1
q − 1
g−1∑j=0
Aj +
2g−2∑j=g
A2g−2qj−g+1
=
qi+1
q − 1DF
(1
q
)− 1
q − 1DF (1).
DUURSMA’S REDUCED POLYNOMIAL 21
According to Lemma 3.1 (i) ⇒ (ii), Duursma’s reduced polynomial of F satis-
fies the functional equation DF (t) = DF
(1qt
)qg−1t2g−2. In particular, DF (1) =
DF
(1q
)qg−1 and there follows (43).
(iii) Due to Ai ≥ 0 for ∀i ≥ 0, Bi are sums of non-negative integers. Moreover,
Bi ≥ Ai(qi+1
q−1
)≥ A0 = 1 > 0 for ∀i ≥ 0 reveals that all Bi are natural numbers.
Towards (44), let us introduce the polynomial ψ(t) :=g−2∑j=0
Ajtj ∈ Z[t] and express
DF (t) =
g−2∑j=0
Ajtj + qg−1t2g−2
g−2∑j=0
Aj(qt)−j+Ag−1tg−1
= ψ(t) + ψ
(1
qt
)qg−1t2g−2 +Ag−1tg−1.
In particular,
DF (1) = ψ(1) + ψ
(1
q
)qg−1 +Ag−1. (52)
Straightforwardly,
Bg−1 − qBg−3
=qg
q − 1
g−2∑j=0
Ajq−j− 1
q − 1
g−2∑j=0
Aj
+Ag−1 −qg−1
q − 1
g−2∑j=0
Ajq−j+
q
q − 1
g−2∑j=0
Aj
= ψ
(1
q
)qg−1 + ψ(1) +Ag−1 = DF (1).
That proves (44) for i = g−1. In the case of g ≤ i ≤ 2g−4 note that 0 ≤ 2g−4−i ≤g − 4 and
(q − 1)(Bi − qi−g+2B2g−4−i)
=
g−1∑j=0
Aj(qi−j+1 − 1) +
i∑j=g
A2g−2−j(qi−g+2 − qj−g+1)−
2g−4−i∑j=0
Aj(qg−1−j − qi−g+2).
Changing the summation index of the second sum to s := 2g − 2− j, one obtains
(q − 1)(Bi − qi−g+2B2g−4−i)
= qi+1
g−1∑j=0
Ajq−j−
g−1∑j=0
Aj
+ qi−g+2
g−2∑s=2g−2−i
As
−qg−1
g−2∑s=2g−2−i
Asq−s− qg−1
2g−4−i∑j=0
Ajq−j+ qi−g+2
2g−4−i∑j=0
Aj
.
22 AZNIV KASPARIAN AND IVAN MARINOV
An appropriate grouping of the sums yields
(q − 1)(Bi − qi−g+2B2g−4−i)
= ψ
(1
q
)qi+1 +Ag−1qi−g+2 − ψ(1)−Ag−1 + ψ(1)qi−g+2 − ψ
(1
q
)qg−1
= (qi−g+2 − 1)
[ψ(1) + ψ
(1
q
)qg−1 +Ag−1
]= DF (1)(qi−g+2 − 1).
That justifies (44).Note that (45) with i ≥ 2g − 2 coincides with (43). In the case of i = 2g − 3,
(q − 1)B2g−3 =
g−1∑j=0
Aj(q2g−2−j − 1) +
g−2∑s=1
As(qg−1 − qg−1−s),
after changing the summation index of the second sum to s := 2g − 2− j. Then
(q − 1)B2g−3
= q2g−2
g−2∑j=0
Ajq−j−
g−2∑j=0
Aj
+Ag−1(qg−1 − 1) + qg−1
g−2∑j=0
Aj
− qg−1g−2∑j=0
Ajq−j
= (qg−1 − 1)
[ψ(1) + ψ
(1
q
)qg−1 +Ag−1
]= DF (1)(qg−1 − 1),
which is tantamount to (45) with i = 2g − 3.(iv) By the Hasse-Weil Theorem, all the roots of LF (t) belong to the circle
S(
1√q
)={z ∈ C | |z| = 1√
q
}. The proof of Proposition 2 specifies that
LF (t) = a2g
g∏j=1
(t− eiϕj√q
)(t− e−iϕj
√q
)
for some ϕj ∈ [0, 2π). The functional equation LF (t) = LF
(1qt
)qgt2g implies that
a2g = qga0. Combining with a0 = LF (0) = 1, one gets
LF (t) =
g∏j=1
(√qt− eiϕj )(√qt− e−iϕj ) =
g∏j=1
(qt2 − 2√q cosϕjt+ 1).
The substitution t = 1 provides
h(F ) = LF (1) =
g∏j=1
(q − 2√q cosϕj + 1).
However, cosϕj ∈ [−1, 1] requires
(√q − 1)2 ≤ q − 2
√q cosϕj + 1 ≤ (
√q + 1)2,
whereas
(√q − 1)2g ≤ h(F ) = LF (1) =
g∏j=1
(q − 2√q cosϕj + 1) ≤ (
√q + 1)2g.
DUURSMA’S REDUCED POLYNOMIAL 23
REFERENCES
[1] S. Dodunekov and I. Landgev, Near MDS-codes, Journal of Geometry, 54 (1995), 30–43.[2] I. Duursma, Weight distribution of geometric Goppa codes, Transections of the American
Mathematical Society, 351 (1999), 3609–3639.
[3] I. Duursma, From weight enumerators to zeta functions, Discrete Applied Mathematics, 111(2001), 55-73.
[4] I. Duursma Combinatorics of the two-variable zeta function in Finite Fields and Applications
Lecture Notes in Computational Sciences 2948, Springer, Berlin, (2004), 109–136.[5] D. Ch. Kim and J. Y. Hyun, A Riemann hypothesis analogue for near-MDS codes, Discrete
Applied Mathematics, 160 (2012), 2440–2444.
[6] H. Niederreiter and Ch. Xing, Algebraic geometry in Coding Theory and Cryptography,Princeton University Press, 2009.
Received xxxx 20xx; revised xxxx 20xx.
E-mail address: [email protected]
E-mail address: [email protected]